I tried with Bayesian optimization for hyperparameters to the support vector regression. I am using 360 data and 4 predictors and one response used 10% holdout for this optimization and am getting low MSE and RMSE if i tuned all 3 parameters (c, gamma, epsilon). but when i used 5-fold cross-validation i tried to optimize the hyperparameters but am not able to get result getting from 10% holdout am getting bit high and its. also later i tried with optimizing 2 parameters by placing one parameter constant at that time am getting some how low MSE i did for both gamma and epsilon values kept constant one at a time and i validate the model. But am not getting what ever the model am getting is right or wrong otherwise the procedure what i am following is right or wrong because am comparing my model with GPR model. am using SVR with gaussian kernel. How i have to validate the model is good or bad. because am getting R-square 100% when am using GPR but when i checked residual plots am bit confused if the model is good or bad. Please any one help me regarding this how i have to validate the model. For epsilon what is the best suitable range and which method hold-out validation or K-fold cross validation is good to develop a model if any one help regarding this. If required i'll post the graphs what am getting. if any support is more helpful to my research. am using matlab for this. Am getting MSE in GPR is 0.0019 and for SVM am getting 4.05 but when am using Optimization technique in different cases am getting different results like 1.35/0.019/0.35/2.46 like this. for this how i have to check my model is fitting properly or if it is overfitting. i am comparing both GPR model and SVR with gaussian kernel. Thanks in advance.

I want to develop the model in microcontroller to validate this model with earlier models to actuate the motor. Actually am trying to predict the motor position by using historical data. So, i want to use this model in any controller and need to check how it will work for predictors to find the position in real time application so for that i collected 360 data set.